WHERE 子句中的函数 - 处理性能不佳

Question

我有一个很大的 table，有超过 100000 条记录。

I need to add one more condition in the where clause used to evalute the totalSum and return only records with the totalSum <> 0.

这里使用了大量的连接和临时表，我并不打算post全部使用。

这是我的函数：

CREATE FUNCTION returnTotalSum(@clientID VARCHAR(20),@type INT,@currency VARCHAR(20),@date VARCHAR())

RETURNS INT
AS
BEGIN

DECLARE @totalSum BIGINT;

SET @totalSum = ( SELECT SUM (CONVERT(DECIMAL(18,4),P.iznos*(1-P.dp)/2))
                   FROM pts as p
                     INNER JOIN tippart t on p.tip = @type
                     INNER JOIN its i on p.partija = @clientID                      
                        WHERE p.currency = @currency and pts.dknizenja < @date
                GROUP BY p.partija )


RETURN @totalSum
END

我在这里使用它：（WHERE 子句中的最后一个 AND）

... 
880001,
NULL, 
NULL,
NULL, 
NULL,
CONVERT(INT,REPLACE('2017.12.31','.',''))                 
            FROM ITS I WITH(NOLOCK)      
              JOIN TIPPART R WITH(NOLOCK) ON I.TIP = R.TIP       
              LEFT JOIN #UNPVT_TIPSTOPE_TS T (NOLOCK) ON I.KAMGRUPA = T.GRUPA AND I.TIP = T.TIP AND I.VALUTA = T.SIFVAL   
              LEFT JOIN #UNPVT_TIPSTOPE_TS T1 (NOLOCK) ON I.KAMGRUPA = T1.GRUPA AND I.TIP = T1.TIP AND I.VALUTA = T1.SIFVAL AND T.GRUPA IS NULL         
              LEFT JOIN #TMP_DODATNA_KS DS (NOLOCK) ON I.PARTIJA = DS.PARTIJA AND I.VALUTA = DS.SIFVAL AND I.KAMGRUPA = DS.GRUPA
              LEFT JOIN #NML_RATE N (NOLOCK)  ON I.TIP = N.TIP AND N.SIFVAL = I.VALUTA AND N.GRUPA = I.KAMGRUPA
             -- LEFT JOIN TIPSTOPE TS (NOLOCK) ON I.TIP = TS.TIP AND TS.GRUPA = I.KAMGRUPA AND TS.SIFVAL = I.VALUTA
              LEFT JOIN #NML_RATE_PERIOD NML (NOLOCK) ON I.TIP = NML.TIP AND I.VALUTA = NML.SIFVAL AND NML.GRUPA = I.KAMGRUPA AND NML.SIFVAL = I.VALUTA
             --WHERE NOT EXISTS (SELECT * FROM [dbo].[IC_INPT_AR_X_INT_RATE_SNPST] WHERE PARTIJA = I.PARTIJA AND VALUTA = I.VALUTA AND APP = 'ST')  
                  WHERE I.DOTVARANJE <= '2017.12.31'
                  AND (T.TIP IS NOT NULL 
                        OR T1.TIP IS NOT NULL 
                        OR DS.PARTIJA IS NOT NULL)
                        AND dbo.returnTotalSum(i.partija,r.tip,t.sifval,i.dotvaranje) <> 0

我假设这个问题是它必须遍历每条记录，比较，评估条件。考虑到 table 中没有索引（我无法添加索引，因为我没有权限）它往往会永远运行。

Is there anything I can do to improve the performance of this function, do you have any suggestions on using something else beside functions and what?

Answer 1

您的函数转换为内联 table 值函数应该接近于此。

CREATE FUNCTION returnTotalSum
(
    @clientID VARCHAR(20)
    ,@type INT
    ,@currency VARCHAR(20)
    ,@date VARCHAR(10) --Don't store dates as strings...
)
RETURNS TABLE AS RETURN

SELECT TotalSum = SUM(CONVERT(DECIMAL(18,4), P.iznos * (1 - P.dp) / 2))
FROM pts as p
INNER JOIN tippart t on p.tip = @type
INNER JOIN its i on p.partija = @clientID                      
WHERE p.currency = @currency 
    and pts.dknizenja < @date
GROUP BY p.partija

Answer 2

在您的实现中调用函数作为过滤的一部分 where 导致巨大的性能消耗，这可能是由于多次索引扫描。

作为一般指导，如果您在没有该功能的情况下也能很好地实现这一点，您应该能够减少它。

没有数据结构和示例数据，很难给你准确的解决方案。

试试下面的代码：

    ... 
    880001,
    NULL, 
    NULL,
    NULL, 
    NULL,
    CONVERT(INT,REPLACE('2017.12.31','.','')) 
    FROM ITS I WITH(NOLOCK)      
        JOIN TIPPART R WITH(NOLOCK) ON I.TIP = R.TIP       
        LEFT JOIN #UNPVT_TIPSTOPE_TS T (NOLOCK) ON I.KAMGRUPA = T.GRUPA AND I.TIP = T.TIP AND I.VALUTA = T.SIFVAL   
        LEFT JOIN #UNPVT_TIPSTOPE_TS T1 (NOLOCK) ON I.KAMGRUPA = T1.GRUPA AND I.TIP = T1.TIP AND I.VALUTA = T1.SIFVAL AND T.GRUPA IS NULL 
        LEFT JOIN #TMP_DODATNA_KS DS (NOLOCK) ON I.PARTIJA = DS.PARTIJA AND I.VALUTA = DS.SIFVAL AND I.KAMGRUPA = DS.GRUPA
        LEFT JOIN #NML_RATE N (NOLOCK)  ON I.TIP = N.TIP AND N.SIFVAL = I.VALUTA AND N.GRUPA = I.KAMGRUPA
        -- LEFT JOIN TIPSTOPE TS (NOLOCK) ON I.TIP = TS.TIP AND TS.GRUPA = I.KAMGRUPA AND TS.SIFVAL = I.VALUTA
        LEFT JOIN #NML_RATE_PERIOD NML (NOLOCK) ON I.TIP = NML.TIP AND I.VALUTA = NML.SIFVAL AND NML.GRUPA = I.KAMGRUPA AND NML.SIFVAL = I.VALUTA
        --WHERE NOT EXISTS (SELECT * FROM [dbo].[IC_INPT_AR_X_INT_RATE_SNPST] WHERE PARTIJA = I.PARTIJA AND VALUTA = I.VALUTA AND APP = 'ST')  
    LEFT OUTER JOIN -- Added this join with same logic from function rather than calling a function. 
    (
        SELECT SUM (CONVERT(DECIMAL(18,4),P.iznos*(1-P.dp)/2)) TotalSum
    FROM pts as p
        INNER JOIN tippart t on p.tip = r.tip
        INNER JOIN its i on p.partija = i.partija     
        WHERE p.currency = t.sifval and pts.dknizenja < i.dotvaranje
        GROUP BY p.partija 
    ) SumTable
    WHERE I.DOTVARANJE <= '2017.12.31'
    AND (T.TIP IS NOT NULL 
        OR T1.TIP IS NOT NULL 
        OR DS.PARTIJA IS NOT NULL)
        AND SumTable.TotalSum <> 0  -- This is similar to your old logic where you were comparing with function output.

查询说明：

我在 SumTable 中添加了与现有查询逻辑的左外连接。
由于附加的左外部联接按 p.partija 分组，因此不会弄乱您的结果集。
您的旧函数的所有输入都已替换为相关值，因为我们在这里进行内联查询。
最后，where 部分被指出是性能消耗的罪魁祸首得到改进，不再调用函数，而是使用 SumTable.TotalSum 并将其与 0 进行比较。

WHERE 子句中的函数 - 处理性能不佳

Function in WHERE clause - dealing with poor performance

sql

sql-server

function

tsql

sql-tuning